Python keras.optimizers.Adam() Examples

def build_model(config):
    """Builds the cnn."""
    params = config.model_arch
    get_model = getattr(models, 'get_model_'+str(params['architecture']))
    model = get_model(params)
    #model = model_kenun.build_convnet_model(params)
    # Learning setup
    t_params = config.training_params
    sgd = SGD(lr=t_params["learning_rate"], decay=t_params["decay"],
              momentum=t_params["momentum"], nesterov=t_params["nesterov"])
    adam = Adam(lr=0.001, beta_1=0.9, beta_2=0.999, epsilon=1e-08)
    optimizer = eval(t_params['optimizer'])
    metrics = ['mean_squared_error']
    if config.model_arch["final_activation"] == 'softmax':
        metrics.append('categorical_accuracy')
    if t_params['loss_func'] == 'cosine':
        loss_func = eval(t_params['loss_func'])
    else:
        loss_func = t_params['loss_func']
    model.compile(loss=loss_func, optimizer=optimizer,metrics=metrics)

    return model 

def train_model():
    if cxl_model:
        embedding_matrix = load_embedding()
    else:
        embedding_matrix = {}
    train, label = vocab_train_label(train_path, vocab=vocab, tags=tag, max_chunk_length=length)
    n = np.array(label, dtype=np.float)
    labels = n.reshape((n.shape[0], n.shape[1], 1))
    model = Sequential([
        Embedding(input_dim=len(vocab), output_dim=300, mask_zero=True, input_length=length, weights=[embedding_matrix],
                  trainable=False),
        SpatialDropout1D(0.2),
        Bidirectional(layer=LSTM(units=150, return_sequences=True, dropout=0.2, recurrent_dropout=0.2)),
        TimeDistributed(Dense(len(tag), activation=relu)),
    ])
    crf_ = CRF(units=len(tag), sparse_target=True)
    model.add(crf_)
    model.compile(optimizer=Adam(), loss=crf_.loss_function, metrics=[crf_.accuracy])
    model.fit(x=np.array(train), y=labels, batch_size=16, epochs=4, callbacks=[RemoteMonitor()])
    model.save(model_path) 

def create_model():
    inputs = Input(shape=(length,), dtype='int32', name='inputs')
    embedding_1 = Embedding(len(vocab), EMBED_DIM, input_length=length, mask_zero=True)(inputs)
    bilstm = Bidirectional(LSTM(EMBED_DIM // 2, return_sequences=True))(embedding_1)
    bilstm_dropout = Dropout(DROPOUT_RATE)(bilstm)
    embedding_2 = Embedding(len(vocab), EMBED_DIM, input_length=length)(inputs)
    con = Conv1D(filters=FILTERS, kernel_size=2 * HALF_WIN_SIZE + 1, padding='same')(embedding_2)
    con_d = Dropout(DROPOUT_RATE)(con)
    dense_con = TimeDistributed(Dense(DENSE_DIM))(con_d)
    rnn_cnn = concatenate([bilstm_dropout, dense_con], axis=2)
    dense = TimeDistributed(Dense(len(chunk_tags)))(rnn_cnn)
    crf = CRF(len(chunk_tags), sparse_target=True)
    crf_output = crf(dense)
    model = Model(input=[inputs], output=[crf_output])
    model.compile(loss=crf.loss_function, optimizer=Adam(), metrics=[crf.accuracy])
    return model 

def evaluate(self, inputs, outputs):
        keras.backend.clear_session()

        X = Input(self.X_train[0].shape)
        co.forward({inputs['in']: X})
        logits = outputs['out'].val
        probs = Activation('softmax')(logits)

        model = Model(inputs=[inputs['in'].val], outputs=[probs])
        model.compile(optimizer=Adam(lr=self.learning_rate),
                      loss='sparse_categorical_crossentropy',
                      metrics=['accuracy'])
        model.summary()
        history = model.fit(self.X_train,
                            self.y_train,
                            batch_size=self.batch_size,
                            epochs=self.num_training_epochs,
                            validation_data=(self.X_val, self.y_val))
        results = {'validation_accuracy': history.history['val_accuracy'][-1]}
        return results 

def fit(self, X, y):
        os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
        assert len(X.shape) == 2
        N, d = X.shape

        from keras.models import Sequential
        from keras.layers import Dense
        from keras.optimizers import Adam
        model = Sequential()
        model.add(Dense(10, input_dim=d, activation="relu"))
        model.add(Dense(10, activation="relu"))
        model.add(Dense(1, activation="relu"))
        model.compile(loss="mse", optimizer=Adam(lr=0.005))
        self.model = model

        n_epochs = 100
        self.model.fit(X, y, epochs=n_epochs, verbose=False) 

def optimizer(self):
        action = K.placeholder(shape=[None, 5])
        discounted_rewards = K.placeholder(shape=[None, ])
        
        # 크로스 엔트로피 오류함수 계산
        action_prob = K.sum(action * self.model.output, axis=1)
        cross_entropy = K.log(action_prob) * discounted_rewards
        loss = -K.sum(cross_entropy)
        
        # 정책신경망을 업데이트하는 훈련함수 생성
        optimizer = Adam(lr=self.learning_rate)
        updates = optimizer.get_updates(self.model.trainable_weights,[],
                                        loss)
        train = K.function([self.model.input, action, discounted_rewards], [],
                           updates=updates)

        return train

    # 정책신경망으로 행동 선택 

def load_model(stamp):
	"""
	"""

	json_file = open(stamp+'.json', 'r')
	loaded_model_json = json_file.read()
	json_file.close()
	model = model_from_json(loaded_model_json, {'AttentionWithContext': AttentionWithContext})

	model.load_weights(stamp+'.h5')
	print("Loaded model from disk")

	model.summary()


	adam = Adam(lr=0.001)
	model.compile(loss='binary_crossentropy',
		optimizer=adam,
		metrics=[f1_score])


	return model 

def optimizer(self):
        action = K.placeholder(shape=[None, 5])
        discounted_rewards = K.placeholder(shape=[None, ])

        # Calculate cross entropy error function
        action_prob = K.sum(action * self.model.output, axis=1)
        cross_entropy = K.log(action_prob) * discounted_rewards
        loss = -K.sum(cross_entropy)

        # create training function
        optimizer = Adam(lr=self.learning_rate)
        updates = optimizer.get_updates(self.model.trainable_weights, [],
                                        loss)
        train = K.function([self.model.input, action, discounted_rewards], [],
                           updates=updates)

        return train

    # get action from policy network 

def actor_optimizer(self):
        action = K.placeholder(shape=(None, self.action_size))
        advantages = K.placeholder(shape=(None, ))

        policy = self.actor.output

        good_prob = K.sum(action * policy, axis=1)
        eligibility = K.log(good_prob + 1e-10) * K.stop_gradient(advantages)
        loss = -K.sum(eligibility)

        entropy = K.sum(policy * K.log(policy + 1e-10), axis=1)

        actor_loss = loss + 0.01*entropy

        optimizer = Adam(lr=self.actor_lr)
        updates = optimizer.get_updates(self.actor.trainable_weights, [], actor_loss)
        train = K.function([self.actor.input, action, advantages], [], updates=updates)
        return train

    # make loss function for Value approximation 

def value_distribution_network(input_shape, num_atoms, action_size, learning_rate):
        """Model Value Distribution

        With States as inputs and output Probability Distributions for all Actions
        """

        state_input = Input(shape=(input_shape)) 
        cnn_feature = Convolution2D(32, 8, 8, subsample=(4,4), activation='relu')(state_input)
        cnn_feature = Convolution2D(64, 4, 4, subsample=(2,2), activation='relu')(cnn_feature)
        cnn_feature = Convolution2D(64, 3, 3, activation='relu')(cnn_feature)
        cnn_feature = Flatten()(cnn_feature)
        cnn_feature = Dense(512, activation='relu')(cnn_feature)

        distribution_list = []
        for i in range(action_size):
            distribution_list.append(Dense(num_atoms, activation='softmax')(cnn_feature))

        model = Model(input=state_input, output=distribution_list)

        adam = Adam(lr=learning_rate)
        model.compile(loss='categorical_crossentropy',optimizer=adam)

        return model 

def optimizer(self):
        action = K.placeholder(shape=[None, 5])
        discounted_rewards = K.placeholder(shape=[None, ])
        
        # 크로스 엔트로피 오류함수 계산
        action_prob = K.sum(action * self.model.output, axis=1)
        cross_entropy = K.log(action_prob) * discounted_rewards
        loss = -K.sum(cross_entropy)
        
        # 정책신경망을 업데이트하는 훈련함수 생성
        optimizer = Adam(lr=self.learning_rate)
        updates = optimizer.get_updates(self.model.trainable_weights,[],
                                        loss)
        train = K.function([self.model.input, action, discounted_rewards], [],
                           updates=updates)

        return train

    # 정책신경망으로 행동 선택 

def _build(self):
        # the model that will be trained
        rnn_x = Input(shape=(None, Z_DIM + ACTION_DIM))
        lstm = LSTM(HIDDEN_UNITS, return_sequences=True, return_state=True)

        lstm_output, _, _ = lstm(rnn_x)
        mdn = Dense(Z_DIM)(lstm_output)

        rnn = Model(rnn_x, mdn)

        # the model used during prediction
        state_input_h = Input(shape=(HIDDEN_UNITS,))
        state_input_c = Input(shape=(HIDDEN_UNITS,))
        state_inputs = [state_input_h, state_input_c]
        
        _, state_h, state_c = lstm(rnn_x, initial_state=state_inputs)
        forward = Model([rnn_x] + state_inputs, [state_h, state_c])

        optimizer = Adam(lr=0.0001)
        # optimizer = SGD(lr=0.0001, decay=1e-4, momentum=0.9, nesterov=True)
        rnn.compile(loss='mean_squared_error', optimizer=optimizer)

        return [rnn, forward] 

def build_3dcnn_model(self, fusion_type, Fusion):
        if len(Fusion[0]) == 1: 
            input_shape = (32, 32, len(Fusion))
            model_in,model = self.cnn_2D(input_shape) 
        else:
            input_shape = (32, 32, 5, len(Fusion))
            model_in,model = self.cnn_3D(input_shape) 
        model = Dropout(0.5)(model)
        model = Dense(32, activation='relu', name = 'fc2')(model)
        model = Dense(self.config.classes, activation='softmax', name = 'fc3')(model) 
        model = Model(input=model_in,output=model)
        # 统计参数
        # model.summary()
        plot_model(model,to_file='experiments/img/' + str(Fusion) + fusion_type + r'_model.png',show_shapes=True)
        print('    Saving model  Architecture')
        
        adam = Adam(lr=0.001, beta_1=0.9, beta_2=0.999, epsilon=1e-8)
        # model.compile(optimizer=adam, loss=self.mycrossentropy, metrics=['accuracy']) #有改善，但不稳定
        model.compile(optimizer=adam, loss='categorical_crossentropy', metrics=['accuracy']) 
        
        return model 

def build_vgg16(image_size=None):
	image_size = image_size or (240, 240)
	if K.image_dim_ordering() == 'th':
	    input_shape = (3,) + image_size
	else:
	    input_shape = image_size + (3, )
	bottleneck_model = vgg16.VGG16(include_top=False, 
	                               input_tensor=Input(input_shape))
	#bottleneck_model.trainable = False
	for layer in bottleneck_model.layers:
	    layer.trainable = False

	x = bottleneck_model.input
	y = bottleneck_model.output
	y = Flatten()(y)
	y = BatchNormalization()(y)
	y = Dense(2048, activation='relu')(y)
	y = Dropout(.5)(y)
	y = Dense(1024, activation='relu')(y)
	y = Dropout(.5)(y)
	y = Dense(1)(y)

	model = Model(input=x, output=y)
	model.compile(optimizer=Adam(lr=1e-4), loss = 'mse')
	return model 

def pre_train_generator(self, g_epochs=3, g_pre_path=None, lr=1e-3):
        if g_pre_path is None:
            self.g_pre_path = os.path.join(self.top, 'data', 'save', 'generator_pre.hdf5')
        else:
            self.g_pre_path = g_pre_path

        g_adam = Adam(lr)
        self.generator_pre.compile(g_adam, 'categorical_crossentropy')
        print('Generator pre-training')
        self.generator_pre.summary()

        self.generator_pre.fit_generator(
            self.g_data,
            steps_per_epoch=None,
            epochs=g_epochs)
        self.generator_pre.save_weights(self.g_pre_path)
        self.reflect_pre_train() 

def siamese_model(type):
  if type=='plate':
    input_shape = (image_size_h_p,image_size_w_p,nchannels)
  else:
    input_shape = (image_size_h_c,image_size_w_c,nchannels)
  left_input = Input(input_shape)
  right_input = Input(input_shape)
  convnet = small_vgg(input_shape)
  encoded_l = convnet(left_input)
  encoded_r = convnet(right_input)

  # Add the distance function to the network
  L1_distance = L1_layer([encoded_l, encoded_r])

  prediction = Dense(2,activation='softmax')(L1_distance)
  optimizer = Adam(0.001, decay=2.5e-4)

  model = Model(inputs=[left_input,right_input],outputs=prediction)
  model.compile(loss="binary_crossentropy",optimizer=optimizer,metrics=['accuracy'])

  return model
#------------------------------------------------------------------------------ 

def __init__(self):
        self.img_rows = 28
        self.img_cols = 28
        self.channels = 1
        self.img_shape = (self.img_rows, self.img_cols, self.channels)
        self.num_classes = 10
        self.latent_dim = 100

        optimizer = Adam(0.0002, 0.5)

        # Build and compile the discriminator
        self.discriminator = self.build_discriminator()
        self.discriminator.compile(
            loss=['binary_crossentropy', 'categorical_crossentropy'],
            loss_weights=[0.5, 0.5],
            optimizer=optimizer,
            metrics=['accuracy']
        )

        # Build the generator
        self.generator = self.build_generator()

        # The generator takes noise as input and generates imgs
        noise = Input(shape=(100,))
        img = self.generator(noise)

        # For the combined model we will only train the generator
        self.discriminator.trainable = False

        # The valid takes generated images as input and determines validity
        valid, _ = self.discriminator(img)

        # The combined model  (stacked generator and discriminator)
        # Trains generator to fool discriminator
        self.combined = Model(noise, valid)
        self.combined.compile(loss=['binary_crossentropy'], optimizer=optimizer) 

def __init__(self):
        self.img_rows = 32
        self.img_cols = 32
        self.mask_height = 8
        self.mask_width = 8
        self.channels = 3
        self.num_classes = 2
        self.img_shape = (self.img_rows, self.img_cols, self.channels)
        self.missing_shape = (self.mask_height, self.mask_width, self.channels)

        optimizer = Adam(0.0002, 0.5)

        # Build and compile the discriminator
        self.discriminator = self.build_discriminator()
        self.discriminator.compile(loss='binary_crossentropy',
            optimizer=optimizer,
            metrics=['accuracy'])

        # Build the generator
        self.generator = self.build_generator()

        # The generator takes noise as input and generates the missing
        # part of the image
        masked_img = Input(shape=self.img_shape)
        gen_missing = self.generator(masked_img)

        # For the combined model we will only train the generator
        self.discriminator.trainable = False

        # The discriminator takes generated images as input and determines
        # if it is generated or if it is a real image
        valid = self.discriminator(gen_missing)

        # The combined model  (stacked generator and discriminator)
        # Trains generator to fool discriminator
        self.combined = Model(masked_img , [gen_missing, valid])
        self.combined.compile(loss=['mse', 'binary_crossentropy'],
            loss_weights=[0.999, 0.001],
            optimizer=optimizer) 

def __init__(self):
        self.img_rows = 32
        self.img_cols = 32
        self.channels = 1
        self.img_shape = (self.img_rows, self.img_cols, self.channels)
        self.mask_height = 10
        self.mask_width = 10
        self.num_classes = 10

        # Number of filters in first layer of generator and discriminator
        self.gf = 32
        self.df = 32

        optimizer = Adam(0.0002, 0.5)

        # Build and compile the discriminator
        self.discriminator = self.build_discriminator()
        self.discriminator.compile(loss=['mse', 'categorical_crossentropy'],
            loss_weights=[0.5, 0.5],
            optimizer=optimizer,
            metrics=['accuracy'])

        # Build the generator
        self.generator = self.build_generator()

        # The generator takes noise as input and generates imgs
        masked_img = Input(shape=self.img_shape)
        gen_img = self.generator(masked_img)

        # For the combined model we will only train the generator
        self.discriminator.trainable = False

        # The valid takes generated images as input and determines validity
        valid, _ = self.discriminator(gen_img)

        # The combined model  (stacked generator and discriminator)
        # Trains the generator to fool the discriminator
        self.combined = Model(masked_img , valid)
        self.combined.compile(loss=['mse'],
            optimizer=optimizer) 

def __init__(self):
        self.img_rows = 28
        self.img_cols = 28
        self.channels = 1
        self.img_shape = (self.img_rows, self.img_cols, self.channels)
        self.latent_dim = 100

        optimizer = Adam(0.0002, 0.5)

        # Build and compile the discriminator
        self.discriminator = self.build_discriminator()
        self.discriminator.compile(loss=['binary_crossentropy'],
            optimizer=optimizer,
            metrics=['accuracy'])

        # Build the generator
        self.generator = self.build_generator()

        # Build the encoder
        self.encoder = self.build_encoder()

        # The part of the bigan that trains the discriminator and encoder
        self.discriminator.trainable = False

        # Generate image from sampled noise
        z = Input(shape=(self.latent_dim, ))
        img_ = self.generator(z)

        # Encode image
        img = Input(shape=self.img_shape)
        z_ = self.encoder(img)

        # Latent -> img is fake, and img -> latent is valid
        fake = self.discriminator([z, img_])
        valid = self.discriminator([z_, img])

        # Set up and compile the combined model
        # Trains generator to fool the discriminator
        self.bigan_generator = Model([z, img], [fake, valid])
        self.bigan_generator.compile(loss=['binary_crossentropy', 'binary_crossentropy'],
            optimizer=optimizer) 

def __init__(self):
        # Input shape
        self.img_rows = 28
        self.img_cols = 28
        self.channels = 1
        self.img_shape = (self.img_rows, self.img_cols, self.channels)
        self.num_classes = 10
        self.latent_dim = 100

        optimizer = Adam(0.0002, 0.5)

        # Build and compile the discriminator
        self.discriminator = self.build_discriminator()
        self.discriminator.compile(loss=['binary_crossentropy'],
            optimizer=optimizer,
            metrics=['accuracy'])

        # Build the generator
        self.generator = self.build_generator()

        # The generator takes noise and the target label as input
        # and generates the corresponding digit of that label
        noise = Input(shape=(self.latent_dim,))
        label = Input(shape=(1,))
        img = self.generator([noise, label])

        # For the combined model we will only train the generator
        self.discriminator.trainable = False

        # The discriminator takes generated image as input and determines validity
        # and the label of that image
        valid = self.discriminator([img, label])

        # The combined model  (stacked generator and discriminator)
        # Trains generator to fool discriminator
        self.combined = Model([noise, label], valid)
        self.combined.compile(loss=['binary_crossentropy'],
            optimizer=optimizer) 

def __init__(self):
        self.img_rows = 28
        self.img_cols = 28
        self.channels = 1
        self.img_shape = (self.img_rows, self.img_cols, self.channels)
        self.latent_dim = 100

        optimizer = Adam(0.0002, 0.5)

        # Build and compile the discriminator
        self.discriminator = self.build_discriminator()
        self.discriminator.compile(loss='mse',
            optimizer=optimizer,
            metrics=['accuracy'])

        # Build the generator
        self.generator = self.build_generator()

        # The generator takes noise as input and generated imgs
        z = Input(shape=(self.latent_dim,))
        img = self.generator(z)

        # For the combined model we will only train the generator
        self.discriminator.trainable = False

        # The valid takes generated images as input and determines validity
        valid = self.discriminator(img)

        # The combined model  (stacked generator and discriminator)
        # Trains generator to fool discriminator
        self.combined = Model(z, valid)
        # (!!!) Optimize w.r.t. MSE loss instead of crossentropy
        self.combined.compile(loss='mse', optimizer=optimizer) 

def __init__(self):
        # Input shape
        self.img_rows = 28
        self.img_cols = 28
        self.channels = 1
        self.img_shape = (self.img_rows, self.img_cols, self.channels)
        self.num_classes = 10
        self.latent_dim = 100

        optimizer = Adam(0.0002, 0.5)
        losses = ['binary_crossentropy', 'sparse_categorical_crossentropy']

        # Build and compile the discriminator
        self.discriminator = self.build_discriminator()
        self.discriminator.compile(loss=losses,
            optimizer=optimizer,
            metrics=['accuracy'])

        # Build the generator
        self.generator = self.build_generator()

        # The generator takes noise and the target label as input
        # and generates the corresponding digit of that label
        noise = Input(shape=(self.latent_dim,))
        label = Input(shape=(1,))
        img = self.generator([noise, label])

        # For the combined model we will only train the generator
        self.discriminator.trainable = False

        # The discriminator takes generated image as input and determines validity
        # and the label of that image
        valid, target_label = self.discriminator(img)

        # The combined model  (stacked generator and discriminator)
        # Trains the generator to fool the discriminator
        self.combined = Model([noise, label], [valid, target_label])
        self.combined.compile(loss=losses,
            optimizer=optimizer) 

def __init__(self):
        self.img_rows = 28
        self.img_cols = 28
        self.channels = 1
        self.img_shape = (self.img_rows, self.img_cols, self.channels)
        self.latent_dim = 100

        optimizer = Adam(0.0002, 0.5)

        # Build and compile the discriminator
        self.d1, self.d2 = self.build_discriminators()
        self.d1.compile(loss='binary_crossentropy',
            optimizer=optimizer,
            metrics=['accuracy'])
        self.d2.compile(loss='binary_crossentropy',
            optimizer=optimizer,
            metrics=['accuracy'])

        # Build the generator
        self.g1, self.g2 = self.build_generators()

        # The generator takes noise as input and generated imgs
        z = Input(shape=(self.latent_dim,))
        img1 = self.g1(z)
        img2 = self.g2(z)

        # For the combined model we will only train the generators
        self.d1.trainable = False
        self.d2.trainable = False

        # The valid takes generated images as input and determines validity
        valid1 = self.d1(img1)
        valid2 = self.d2(img2)

        # The combined model  (stacked generators and discriminators)
        # Trains generators to fool discriminators
        self.combined = Model(z, [valid1, valid2])
        self.combined.compile(loss=['binary_crossentropy', 'binary_crossentropy'],
                                    optimizer=optimizer) 

def __init__(self):
        # Input shape
        self.img_rows = 28
        self.img_cols = 28
        self.channels = 1
        self.img_shape = (self.img_rows, self.img_cols, self.channels)
        self.latent_dim = 100

        optimizer = Adam(0.0002, 0.5)

        # Build and compile the discriminator
        self.discriminator = self.build_discriminator()
        self.discriminator.compile(loss='binary_crossentropy',
            optimizer=optimizer,
            metrics=['accuracy'])

        # Build the generator
        self.generator = self.build_generator()

        # The generator takes noise as input and generates imgs
        z = Input(shape=(self.latent_dim,))
        img = self.generator(z)

        # For the combined model we will only train the generator
        self.discriminator.trainable = False

        # The discriminator takes generated images as input and determines validity
        valid = self.discriminator(img)

        # The combined model  (stacked generator and discriminator)
        # Trains the generator to fool the discriminator
        self.combined = Model(z, valid)
        self.combined.compile(loss='binary_crossentropy', optimizer=optimizer) 

def __init__(self):
        self.img_rows = 28
        self.img_cols = 28
        self.channels = 1
        self.img_shape = (self.img_rows, self.img_cols, self.channels)
        self.latent_dim = 100

        optimizer = Adam(0.0002, 0.5)

        # Build and compile the discriminator
        self.discriminator = self.build_discriminator()
        self.discriminator.compile(loss='binary_crossentropy',
            optimizer=optimizer,
            metrics=['accuracy'])

        # Build the generator
        self.generator = self.build_generator()

        # The generator takes noise as input and generates imgs
        z = Input(shape=(self.latent_dim,))
        img = self.generator(z)

        # For the combined model we will only train the generator
        self.discriminator.trainable = False

        # The discriminator takes generated images as input and determines validity
        validity = self.discriminator(img)

        # The combined model  (stacked generator and discriminator)
        # Trains the generator to fool the discriminator
        self.combined = Model(z, validity)
        self.combined.compile(loss='binary_crossentropy', optimizer=optimizer) 

def __init__(self):
        self.img_rows = 28
        self.img_cols = 28
        self.channels = 1
        self.img_shape = (self.img_rows, self.img_cols, self.channels)
        self.latent_dim = 10

        optimizer = Adam(0.0002, 0.5)

        # Build and compile the discriminator
        self.discriminator = self.build_discriminator()
        self.discriminator.compile(loss='binary_crossentropy',
            optimizer=optimizer,
            metrics=['accuracy'])

        # Build the encoder / decoder
        self.encoder = self.build_encoder()
        self.decoder = self.build_decoder()

        img = Input(shape=self.img_shape)
        # The generator takes the image, encodes it and reconstructs it
        # from the encoding
        encoded_repr = self.encoder(img)
        reconstructed_img = self.decoder(encoded_repr)

        # For the adversarial_autoencoder model we will only train the generator
        self.discriminator.trainable = False

        # The discriminator determines validity of the encoding
        validity = self.discriminator(encoded_repr)

        # The adversarial_autoencoder model  (stacked generator and discriminator)
        self.adversarial_autoencoder = Model(img, [reconstructed_img, validity])
        self.adversarial_autoencoder.compile(loss=['mse', 'binary_crossentropy'],
            loss_weights=[0.999, 0.001],
            optimizer=optimizer) 

def __init__(self):
        self.img_rows = 28
        self.img_cols = 28
        self.channels = 1
        self.img_shape = (self.img_rows, self.img_cols, self.channels)
        self.latent_dim = 100

        optimizer = Adam(0.0002, 0.5)

        # Build and compile the discriminator
        self.discriminator = self.build_discriminator()
        self.discriminator.compile(loss='binary_crossentropy',
            optimizer=optimizer,
            metrics=['accuracy'])

        # Build the generator
        self.generator = self.build_generator()

        # The generator takes noise as input and generated imgs
        z = Input(shape=(self.latent_dim,))
        img = self.generator(z)

        # For the combined model we will only train the generator
        self.discriminator.trainable = False

        # The valid takes generated images as input and determines validity
        valid = self.discriminator(img)

        # The combined model  (stacked generator and discriminator)
        # Trains the generator to fool the discriminator
        self.combined = Model(z, valid)
        self.combined.compile(loss=self.boundary_loss, optimizer=optimizer) 

def main(rootdir, case, results):
    train_x, train_y, valid_x, valid_y, test_x, test_y = get_data(args.dataset, case)

    input_shape = (train_x.shape[1], train_x.shape[2])
    num_class = train_y.shape[1]
    if not os.path.exists(rootdir):
        os.makedirs(rootdir)
    filepath = os.path.join(rootdir, str(case) + '.hdf5')
    saveto = os.path.join(rootdir, str(case) + '.csv')
    optimizer = Adam(lr=args.lr, clipnorm=args.clip)
    pred_dir = os.path.join(rootdir, str(case) + '_pred.txt')

    if args.train:
        model = creat_model(input_shape, num_class)
        early_stop = EarlyStopping(monitor='val_acc', patience=15, mode='auto')
        reduce_lr = ReduceLROnPlateau(monitor='val_acc', factor=0.1, patience=5, mode='auto', cooldown=3., verbose=1)
        checkpoint = ModelCheckpoint(filepath, monitor='val_acc', verbose=1, save_best_only=True, mode='auto')
        csv_logger = CSVLogger(saveto)
        if args.dataset=='NTU' or args.dataset == 'PKU':
            callbacks_list = [csv_logger, checkpoint, early_stop, reduce_lr]
        else:
            callbacks_list = [csv_logger, checkpoint]

        model.compile(loss='categorical_crossentropy', optimizer=optimizer, metrics=['accuracy'])
        model.fit(train_x, train_y, validation_data=[valid_x, valid_y], epochs=args.epochs,
                  batch_size=args.batch_size, callbacks=callbacks_list, verbose=2)

    # test
    model = creat_model(input_shape, num_class)
    model.load_weights(filepath)
    model.compile(loss='categorical_crossentropy', optimizer=optimizer, metrics=['accuracy'])

    scores = get_activation(model, test_x, test_y, pred_dir, VA=10, par=9)

    results.append(round(scores, 2)) 

def cnn_model():
    (x_train, y_train), _ = mnist.load_data()
    # 归一化
    x_train = x_train.reshape(-1, 28, 28, 1) / 255.
    # one-hot
    y_train = np_utils.to_categorical(y=y_train, num_classes=10)

    model = Sequential([
        # input_shape:输入平面，就在第一个位置设置
        # filters：卷积核、滤波器
        # kernel_size：卷积核大小
        # strides：步长
        # padding有两种方式：same/valid
        # activation：激活函数
        Convolution2D(input_shape=(28, 28, 1), filters=32, kernel_size=5, strides=1, padding='same', activation=relu),
        MaxPool2D(pool_size=2, strides=2, padding='same'),
        Convolution2D(filters=64, kernel_size=5, padding='same', activation=relu),
        MaxPool2D(pool_size=2, trainable=2, padding='same'),
        Flatten(),  # 扁平化
        Dense(units=1024, activation=relu),
        Dropout(0.5),
        Dense(units=10, activation=softmax),
    ])
    opt = Adam(lr=1e-4)
    model.compile(optimizer=opt, loss=categorical_crossentropy, metrics=['accuracy'])
    model.fit(x=x_train, y=y_train, batch_size=64, epochs=20, callbacks=[RemoteMonitor()])
    model_save(model, './model.h5')